List your company

Text-to-Image Generation on MS-COCO 30k (val)

4.31FIDl-DeTok
1.366421.235741.10560.9743May 24, 2023Oct 2, 2023Feb 11, 2024Jun 21, 2024Oct 31, 2024Mar 11, 2025Jul 21, 2025

Evaluation results

MethodLinks
l-DeTokTokenizer=l-DeTok, Base Generative Model=SiT-B, CFG scale=best2025.074.31—24.61
l-DeTokTokenizer=l-DeTok, Base Generative Model=MAR-B, CFG scale=best2025.074.97—24.82
MAR-VAETokenizer=MAR-VAE, Base Generative Model=SiT-B, CFG scale=best2025.075.74—23.5
VA-VAETokenizer=VA-VAE, Base Generative Model=SiT-B, CFG scale=best2025.075.83—25.07
SD-VAETokenizer=SD-VAE, Base Generative Model=SiT-B, CFG scale=best2025.076.63—24
Stable Diffusion v1.5Guidance Scale=2, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.059.740.3—
MULTIFUSIONGuidance Scale=4, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.059.90.29—
Stable Diffusion v1.5Guidance Scale=4, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0510.190.31—
Efficient UNet + ClockworkResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=5.92023.1211.140.29—
MULTIFUSIONGuidance Scale=6, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0512.150.29—
MULTIFUSIONGuidance Scale=2, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0512.210.28—
SD UNet + ClockworkResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=7.32023.1212.270.291—
Efficient UNetResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=9.52023.1212.330.296—
Distilled Efficient UNet + ClockworkResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=2.92023.1212.370.291—
MAR-VAETokenizer=MAR-VAE, Base Generative Model=MAR-B, CFG scale=best2025.0712.49—22.07
Stable Diffusion v1.5Guidance Scale=6, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0512.730.31—
SD UNetResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=10.82023.1212.770.296—
DUObeta=10002024.0713.370.3078—
SDv1.5 (Teacher)Steps=50, Latency=2.59s, Guidance scale=8, Batch size=12023.1113.450.322—
SD (1.4v)version=1.4v2024.0713.520.3095—
SPM2024.0713.530.3093—
Distilled Efficient UNetResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=4.72023.1213.920.292—
SnapFusionResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=42023.12140.3—
UCE2024.0714.040.3074—
MULTIFUSIONGuidance Scale=8, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0514.190.3—
Stable Diffusion v1.5Guidance Scale=8, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0514.620.31—
DMDSteps=1, Latency=0.09s, Guidance scale=8, Batch size=12023.1114.930.32—
ESD2024.0716.870.2944—
BK-SDM-BaseResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=8.42023.1217.230.287—
BK-SDM-SmallResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=8.22023.1217.720.268—
DUObeta=5002024.0718.280.3018—
BK-SDM-TinyResolution=256x256, Sampling steps=8, Scheduler=DPM++, TFLOPS=7.82023.1218.640.265—
SD-VAETokenizer=SD-VAE, Base Generative Model=MAR-B, CFG scale=best2025.0719.75—22.86
InstaFlowResolution=256x256, Sampling steps=1, Scheduler=None, TFLOPS=0.82023.1220——
DPM++Steps=4, Latency=0.26s, Guidance scale=8, Batch size=12023.1122.440.309—
UniPCSteps=4, Latency=0.26s, Guidance scale=8, Batch size=12023.1123.30.308—
LCM-LoRASteps=4, Latency=0.19s, Guidance scale=7.5, Batch size=12023.1123.620.297—
LCM-LoRASteps=2, Latency=0.12s, Guidance scale=7.5, Batch size=12023.1124.280.294—
Stable Diffusion v1.5Guidance Scale=1, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0526.090.27—
MULTIFUSIONGuidance Scale=1, Prompt Type=Textual, Image Size=256x256, Diffusion Steps=1002023.0532.810.25—
VA-VAETokenizer=VA-VAE, Base Generative Model=MAR-B, CFG scale=best2025.0734.64—21.98
LCM-LoRASteps=1, Latency=0.09s, Guidance scale=7.5, Batch size=12023.1177.90.238—
DPM-Solver++(2M)NFE=8, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——28.5
DPM-Solver++(2M)NFE=12, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——28.65
DPM-Solver++(2M)NFE=16, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——29.65
DPM-Solver++(2M)NFE=20, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——30.02
DyWeightNFE=8, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——28.56
DyWeightNFE=12, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——29.81
DyWeightNFE=16, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——30.17
DyWeightNFE=20, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——30.05
iPNDM(2M)NFE=8, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——28.33
iPNDM(2M)NFE=12, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——29.79
iPNDM(2M)NFE=16, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——30.15
iPNDM(2M)NFE=20, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——30.3
UniPC(2M)NFE=8, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——26.44
UniPC(2M)NFE=12, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——29.05
UniPC(2M)NFE=16, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——29.79
UniPC(2M)NFE=20, Base Model=Stable-Diffusion-v1.5, CFG (Classifier-Free Guidance)=True2026.03——30.08